TY - JOUR
T1 - Reinforcement Learning-Based Prediction of Alarm Significance in Marginally Operating Electrical Grids
AU - Mirshekali, Hamid
AU - Shaker, Hamid Reza
PY - 2024/4
Y1 - 2024/4
N2 - The increasing electrical load strains the grid, leading to frequent overload alarms and reduced reliability. Moreover, as the electrical grid expands, its maintenance poses greater challenges, underscoring the necessity for smart investment strategies. Rather than solely focusing on expanding the grid infrastructure, an alternative approach involves utilizing software-based solutions, which result in predictive and automated maintenance systems. Conversely, due to resource constraints, it becomes crucial to assess the significance of network alarms to prioritize resource allocation effectively. In this article, a novel predictive framework is proposed to predict the significance of overload alarms in marginal operating electrical grids, utilizing the existing infrastructure. A simple clustering model is first introduced which enhance the scalability of the proposed framework. Convolutional long short-term memory, convolutional neural network, recurrent neural network, and some more deep learning models and double deep Q-network and Q-learning techniques are employed for prediction task. Following that, virtual thresholds are defined, wherein exceeding the load results in an overload alarm being raised. Chernoff-Hoeffding bound is employed to assign significance quantifying to the predicted deviations. Subsequently, expert observation of alarms, is input into the particle swarm optimization algorithm to fine-tune parameters for the Chernoff-Hoeffding bound. The effectiveness of the proposed method is evaluated on the residential and industrial loads within a part of the Danish distribution network. The results showcase the effectiveness and robustness of the proposed method in contrast with its other counterparts.
AB - The increasing electrical load strains the grid, leading to frequent overload alarms and reduced reliability. Moreover, as the electrical grid expands, its maintenance poses greater challenges, underscoring the necessity for smart investment strategies. Rather than solely focusing on expanding the grid infrastructure, an alternative approach involves utilizing software-based solutions, which result in predictive and automated maintenance systems. Conversely, due to resource constraints, it becomes crucial to assess the significance of network alarms to prioritize resource allocation effectively. In this article, a novel predictive framework is proposed to predict the significance of overload alarms in marginal operating electrical grids, utilizing the existing infrastructure. A simple clustering model is first introduced which enhance the scalability of the proposed framework. Convolutional long short-term memory, convolutional neural network, recurrent neural network, and some more deep learning models and double deep Q-network and Q-learning techniques are employed for prediction task. Following that, virtual thresholds are defined, wherein exceeding the load results in an overload alarm being raised. Chernoff-Hoeffding bound is employed to assign significance quantifying to the predicted deviations. Subsequently, expert observation of alarms, is input into the particle swarm optimization algorithm to fine-tune parameters for the Chernoff-Hoeffding bound. The effectiveness of the proposed method is evaluated on the residential and industrial loads within a part of the Danish distribution network. The results showcase the effectiveness and robustness of the proposed method in contrast with its other counterparts.
KW - Chernoff–Hoeffding bound
KW - Convolutional neural networks
KW - Data models
KW - Feature extraction
KW - Load modeling
KW - Prediction algorithms
KW - Predictive models
KW - Q-learning
KW - Task analysis
KW - double deep Q-network (DDQN)
KW - marginal grid operation
KW - overload alarm prediction
KW - Chernoffa-Hoeffding bound
U2 - 10.1109/TII.2023.3348819
DO - 10.1109/TII.2023.3348819
M3 - Journal article
SN - 1551-3203
VL - 20
SP - 6510
EP - 6521
JO - IEEE Transactions on Industrial Informatics
JF - IEEE Transactions on Industrial Informatics
IS - 4
ER -